Upload
tranque
View
216
Download
0
Embed Size (px)
Citation preview
1
Basel Committee on Banking Supervision Consultative Document
Fundamental Review of the Trading Book
Dated May 2012
Response of the International Swaps and Derivatives Association, Inc,
the Global Financial Markets Association,
the Institute of International Finance and
the International Banking Federation.
7 September 2012
2
Contents1Introduction ........................................................................................................................................ 4
2 Trading Book Banking Book Boundary ................................................................................................. 7
2.1Introduction ................................................................................................................................. 7
2.2ValuationBasedApproach ......................................................................................................... 8
2.3TradingEvidenceBasedApproach ............................................................................................ 8
2.4Moredetailedpoints ................................................................................................................. 10
2.4.1TheValuationBasedApproachhasSeveralIssues: ........................................................ 10
2.4.2IssueswiththeTradingEvidenceBasedApproach ........................................................ 11
3InterestRateRiskintheBankingBook .......................................................................................... 14
4ChoiceofRiskMetricandCalibrationtoStressedConditions ..................................................... 16
4.1Introduction ............................................................................................................................... 16
4.2AppropriatePercentile ............................................................................................................. 16
4.3DirectversusIndirectApproach .............................................................................................. 17
5LiquidityAdjustedValueatRiskorExpectedShortfall ................................................................ 19
5.1Introduction ............................................................................................................................... 19
5.2PracticalConsiderations ........................................................................................................... 19
5.3ConceptualFramework ............................................................................................................ 20
6 Hedging and Diversification ............................................................................................................... 22
6.1Introduction ............................................................................................................................... 22
6.2Technicalissueswiththeproposedapproach ........................................................................ 22
6.3ProposedApproach,shouldmodelingnotbeacceptable ...................................................... 23
6.4DetailedexamplesonHedging&Diversification ................................................................... 23
7RevisedStandardisedApproach ..................................................................................................... 26
7.1Introduction ............................................................................................................................... 26
7.2MandatoryCalculationofStandardisedCapitalRequirements ............................................ 26
7.3RegulatoryCapitalFloorsbasedontheStandardisedApproach .......................................... 27
8RelationshipbetweenStandardisedandModelsBasedApproaches .......................................... 28
8.1Introduction ............................................................................................................................... 28
8.2 Approval Criteria ......................................................................................................................... 28
8.3 The Treatment of Unapproved Desks or Positions ..................................................................... 28
8.4 A Potential Approach Using Scaling ............................................................................................ 29
8.5 Eligibility for internal models approach ...................................................................................... 30
9CreditRiskintheTradingBook(ExcludingCVA) ......................................................................... 31
3
9.1Introduction ............................................................................................................................... 31
10PartialandFullerRiskFactorApproaches .................................................................................. 33
10.1PartialRiskFactorApproach ................................................................................................. 33
10.2FullerRiskFactorApproach .................................................................................................. 33
10.3Internalmodels‐basedapproaches ....................................................................................... 33
Appendix1:ExpectedShortfallandVaREquivalentsundertheGeneralisedParetoDistribution .............................................................................................................................................................. 35
4
1Introduction The Associations are supportive of the Fundamental Review’s aims and objectives both in strengthening capital standards and delivering a regulatory framework which achieves comparable levels of capital internationally. In discussing the Fundamental Review and formulating our response we have focused on the nature and purpose of capital. In consequence this response should be read in conjunction with previous input including ISDA’s November 2011 paper1. Need for a coherent, risk‐based framework. The Associations are supportive of BCBS’s agenda of reforming bank regulatory standards to address the lessons of the financial crisis. In particular, the Associations believe that it is important to develop a coherent and comprehensive framework which is risk sensitive at both the individual position and portfolio levels. The BCBS has recognised that the principal criticism of the Basel 2.5 framework is the “patchwork” nature of rules. Value at Risk (“VaR”), stressed VaR, Incremental Risk Charge and the Comprehensive Risk Measure all, to some extent, cover the same risks. Further, the Basel I counterparty risk rules capitalise the risk of default whereas the Credit Valuation Adjustment (“CVA”) rules capitalise the risks of credit migration up to and including default, thereby introducing further double counting which magnifies the divergence between economic risk and regulatory capital. In crafting a new coherent framework, great care needs to be taken not only to simplify the overall trading book framework but, more importantly, to achieve a closer alignment between risk and regulatory capital. Any new regulatory framework that will be introduced must be clearly aligned to the economic risks of the business. Where firms are forced to carry out calculations which are not central to managing the risks, these are unlikely to prove a good use of resource. More importantly, trading is an area where solutions that are too simple or eschew modern modeling capabilities in a simplistic way can also cause systemic issues and unfortunate effects not only on the bank’s business but also on the market and on the economy. This is a very important point especially now that regulatory capital is rapidly becoming a significant constraint on the business. It follows that a failure to more effectively align risk and regulatory capital will distort the effective deployment of capital and affect business decisions with often adverse economic consequences. In this respect, it is important to recognise that firms’ internal models have been significantly upgraded since the financial crisis and leverage has been reduced. There has also been a large reduction in risk following the introduction of Basel 2.5. Firms should be permitted to utilize these significantly enhanced models.
The Associations strongly support the continued development of risk sensitive models to calculate regulatory capital. Firms should be encouraged to continually develop and improve models, and work with regulators to strengthen modeling standards. This contrasts with certain aspects of the Basel III rules where there appear to be disincentives to modeling.
Need for a framework that would be reasonably comparable across jurisdictions. The international dimension is clearly important and the BCBS recognises that, unless agreement is reached across jurisdictions, the resulting “un‐level playing field” may pose significant threats to financial stability. We fully support this aim and urge regulators to agree international standards whilst allowing
1 The Marker Risk Capital Framework, A contribution to the Fundamental Review of the Trading Book, November 2011
5
sufficient flexibility for local implementation where necessary to reflect market specificities within a coherent, risk‐based, overall approach. This is not easy and it will require hard work to ensure strong standards are applied in a reasonably comparable manner. But it is essential.
Need to take account of strong governance and internal controls, sound risk management, and effective supervision in the fundamental review. Minimum capital standards are just one component of an effective regulatory regime and have been already materially increased under Basel 2.5 and Basel 3. Individual jurisdictions have in addition authority to impose capital add‐ons above those minima and many are in the process of implementing “gold plated” versions of the international standards (for example in Europe with systemic buffer and the macro prudential measures). Risk management standards, management information systems and senior level awareness of the business and its inherent risks are also critical and it is vital that the Fundamental Review put as much emphasis on the importance of governance, risk management, independent control and effective supervision as on minimum capital requirements. The IIF‐Ernst & Young paper – Progress in financial services risk management (June 2012) – provide some evidence of progress in these areas.
Trading Book/Banking Book boundary. We broadly support the trading evidence‐based boundary because it aligns more closely regulatory capital with how risks are managed by banks. In addition, it should be noted that the concept of portfolio is important for the Trading Book/Banking Book Boundary question. We believe that the unit of account in determining the Boundary should be at the portfolio, and not at the position, level. It is important that portfolios are not split between the Trading and Banking Books as the breaking up of netting sets and separation of hedges from underlying positions can have serious unintended consequences.
Market liquidity. We support the idea to incorporate market liquidity in the trading book capital
framework. As we have learned from the recent financial crisis, a sudden and severe impairment of
liquidity can lead to difficulties in hedging and exiting positions, resulting in significant cumulative
mark‐to‐market losses. However, this should be done carefully to minimize unintended
consequences.
Model approval. We are supportive of the change in model approval to operate at the desk level. However, the term “desk” needs further discussion and steps need to be taken to ensure that portfolios are “in” or “out” to ensure that positions are not split from their associated hedges. We also put forward a proposal to smooth the transition from model to non‐model approaches to eliminate the “cliff effect” of switching off model approval.
Relationship between standardised and internal model approaches. There should be coherence between the standardised and internal model approaches with the former being enhanced to be more risk sensitive than the current standardised rules. We recognize, however, that there are likely to be issues applying complex standardised methodologies to smaller, less sophisticated banks. Perhaps some simpler, and sufficiently conservative, standardised rules should be developed for them.
Implementation of both standard and internal approaches (“IMA”) will require a substantial investment from firms both at inception (e.g., feeding all relevant information into the regulatory calculation process) and for regular production (market data, mapping, maintenance etc.). Therefore, the operational consequences should be carefully considered when the TBG assesses the frequency of how often IMA firms need to calculate the standard approach capital figures.
6
Proposed revised standardised approaches. We believe that the Fuller Risk Factor Approach is closer to the models currently used by firms under the internal models approach of the trading book framework, and will likely be chosen by them. For firms with no current model approval, developing a Partial Risk Factor Approach will involve almost as much effort as developing a Fuller Risk Factor Approach Model but without yielding many of the advantages.
CVA. We believe that the Basel III proposals for Credit Valuation Adjustment (“ CVA”) require further development. The Basel 3 framework achieves higher levels of capital, but the distribution of that capital burden has already begun to skew market practice and marginalized the use of derivatives in certain activities unrelated to the financial crisis. We are keen to work with the appropriate group within the Basel Committee to produce a more coherent and workable CVA framework that is aligned to the objectives of the Fundamental Review. We will follow this response with a paper more focused on CVA issues. However, it should be clear to the Committee that if they leave the CVA variability charge untouched whilst changing the Trading Book rules, new regulatory arbitrage opportunities will open up between the CVA hedge book and the trading book. Careful reconsideration of the where CVA risk lies relative to the Trading Book/Banking Book boundary should be the starting point of an extensive review.
QIS. The proposed Quantitative Impact Study needs close industry involvement and we remain committed to working with regulators in this important area. The QIS should aim not only to test the detailed proposals that the TBG will issue, but to determine their macroprudential as well as microprudential implications.
7
2TradingBookBankingBookBoundary
2.1Introduction
In relation to Question 1 in section 3.1 of the Fundamental Review:
“Which Boundary Option do you believe would best address the weaknesses identified with the
current boundary, whilst meeting the Committee’s objectives?”
For the Trading Book/Banking Book Boundary (the “Boundary”) we broadly support the first option
set out in the Fundamental Review; that of a trading evidence based boundary because it aligns
more closely regulatory capital with how risks are managed by banks. However, we believe that the
boundary should be defined at the portfolio, and not the position, level for further commentary see
ISDA’s June 2011 paper2.
We believe that the Fundamental Review should begin by defining the function and characteristics
of each Book i.e. Banking and Trading and then determining the appropriate minimum level of
capital required. This should include analysis of the nature of the risks and how they are managed, as
well as the role and purpose of capital which was discussed in a paper by ISDA published in
November 20113. . We would like to work with the BCBS on this.
By way of example, during the recent financial crisis, very large economic losses occurred to credit
sensitive instruments in the trading book, primarily driven by a dramatic increase in the liquidity risk
premium. For many types of credit sensitive products, an objective assessment of the stressed
potential cumulative lifetime credit losses of the portfolio was less than the loss in price. These
portfolios would thus require more capital in a trading book than a banking book, because the
stressed price loss of the former exceeded the stressed cumulative lifetime default loss of the latter.
The conclusion from this observation is that aligning minimum capital requirements to risk does not
necessarily mean that the capital required for a portfolio of credit sensitive instruments needs to be
the same in the banking book and the trading book. Different capital requirements can be
appropriate, if the nature of the risk and how the risk is managed are different.
BCBS has consulted on the “permeability” of the Boundary. Whilst broadly in agreement that
migration between the Trading and Banking Books should be rare and closely controlled, we believe
that the Boundary should not be impermeable as trading intent, and hence the evidence supporting
that intent, can change.
Although not expressed in FRTB, we believe the purpose of market risk capital in the Trading Book is
to provide a cushion to absorb losses to avoid insolvency. Losses arise from positions being marked
to market or “fair valued”. At first sight, therefore, the valuation based approach seems more
2 Notes on the Trading Book/Banking Book Boundary June 2011 3 The Market Risk Capital Framework: A Contribution to the Fundamental Review of the Trading Book. ISDA November 2011.
8
coherent. However we have looked closely at this approach and set out below the reasons for our
preference. These revolve principally around the inclusion into the Trading Book of portfolios
currently and appropriately held in the Banking Book but which are fair valued.
2.2ValuationBasedApproach
The Valuation‐based approach, as written, has several issues that make it undesirable from a policy
perspective. It represents a significant change from current practices which are well understood and
have incorporated lessons from the financial crisis. Essentially the way in which it is formulated
requires the assets (but not the liabilities) of Available for Sale (“AFS”) and other portfolios which are
fair valued to be included in the Trading Book and assigned market Risk Weighted Assets (“RWA's”).
Although certain carve‐outs are contemplated, the effect of this is:
It creates a material break between the regulatory categorization of risk and how the
risk is actually managed.
It further creates a material break between economic risk and RWA’s, in that the liability
side of the balance sheet is ignored in the latter case.
Further, the proposal may result in significant inconsistencies in RWA’s due to differences in
accounting standards between jurisdictions. As set out in the June ISDA paper: “ It is more
appropriate for bank regulation to have an appropriate boundary than for it to have one that is
completely consistent with accounting, especially given that accounting is not itself internally
consistent.”
We note that although the Basel Committee defined the scope of its application by an “intent to trade” standard, the US implementation of the 1996 Market Risk Amendment (“MRA”) defined the scope of its application by accounting category. In the US, the MRA only applied to instruments in the Trading category albeit some such positions do not qualify for regulatory trading book treatment. This is an example of an accounting definition of trading that does not include assets in an Available
For Sale (“AFS”) or a loan portfolio.
2.3TradingEvidenceBasedApproach
The Trading Evidence Based Approach aligns regulatory capital with how risks are managed by banks. It is well understood by senior management and regulators. Risk management for trading books tends to be carried out on a very proactive basis, typically daily, and sometimes intra‐day. This includes limits which are set and monitored for appropriateness, daily marking‐to‐market, portfolios which are reported to senior management as an integral part of the institution’s risk management process, active and observable management action and rehedging/rebalancing of risks. Risk management occurs at the portfolio level and therefore the trading book classification should also occur at this level.
9
This approach also overcomes the other material concerns with the Valuation Based Approach and
presents a workable solution. In particular it eliminates the break between the way in which fair
valued portfolios in the Banking Book are managed and their accounting treatment. However, there
remain a small number of shortcomings which we summarise below but cover in greater detail in
Section 2.4.2:
We would like to work closely with the BCBS to arrive at the “objective evidential
requirements” to substantiate trading intent, and wish to highlight from the outset that an
evidence based framework should act as guidance rather than a rigid checkbox exercise,
reflecting the fact that trading intent cannot be evidenced in every case in exactly the same
way.
The Industry is concerned with the difficulties in assessing the feasibility of trading liquidity
(see section 5).
The rules for transferring assets from the Trading Book to the Banking Book (permeability)
should not be so rigid as to prevent a bank from making that switch, so long as there is both
strong supervisory oversight and bank governance and controls that monitor these
movements. This may occur if market liquidity dries up and the bank chooses to hold to
maturity rather than continue to hold the assets in a trading portfolio. If trading intent
changes due to exogenous conditions then the evidence supporting that intent will also
change.
The BCBS has highlighted the concern that certain fair valued assets classified in the banking
book will not receive appropriate prudential requirements. We understand and agree with
this concern. We therefore acknowledge that further work needs to be undertaken to
properly capture the risks in such portfolios. We note that the credit risk of AFS portfolios
(both bonds and equities) is already addressed in the existing regulatory framework.
(Interest Rate Risk in the Banking Book or “IRRBB”) is currently captured by a Pillar 2 context.
Including IRRBB under Pillar 1 is not easy, because many of the assets and liabilities in the
banking book have indefinite maturities. The effective tenors assigned to such assets and
liabilities will have a material impact on the measured interest rate risk. The Associations
are willing to work closely with the BCBS on the feasibility of establishing Pillar 1 standards
for measuring IRRBB. We believe this will likely lead to a more appropriate outcome than
simply classifying such portfolios as trading, when there is a clear lack of trading intent, and
where the portfolios in question are not risk managed like other trading portfolios.
10
2.4Moredetailedpoints
2.4.1TheValuationBasedApproachhasSeveralIssues:
TheValuationBasedApproachcausesamaterialbreakbetweentheregulatorycategorizationofriskandhowtheriskisactuallymanaged:
This proposal may very materially expands the set of instruments and portfolios included in the
calculation of RWA’s for trading, even though those newly captured portfolios are not currently
managed like trading portfolios. The proposal would affect both debt and equity securities held in
AFS portfolios. Most fixed income portfolios categorized as AFS are held for the purpose of liquidity
risk management (including investing excess liquidity), and/or because local regulators require banks
to hold local currency sovereign debt securities. These are typically long‐term structured positions
that are not managed like trading portfolios.
TheValuationBasedApproachcausesamaterialbreakbetweeneconomicriskandriskweightedassets:
The proposal would apply RWA methodology to AFS portfolios as an artefact of AFS accounting,
without any relationship to the actual interest rate risk of such portfolios. This would capture the
market risk of the AFS portfolio assets but not the offsetting risk of the liabilities that fund these
assets. The interest rate risk of the AFS portfolio, like other banking book portfolios, may also
include interest rate derivative hedges. The actual measure of the economic risk of the portfolio
requires the inclusion of all relevant assets, liabilities, and hedges, although certain carve‐outs are
contemplated for positions that hedge interest rate risk in the banking book.
An extreme example, to illustrate the problem, would be an AFS portfolio that was a “matched
book”, with no interest rate risk. Because of the unilateral focus on the market risk of the assets,
independent of the liabilities which fund them, the proposal would result in a large marginal
increase in Trading Book RWA, even though by construction the AFS portfolio in this example had no
interest rate risk. The interest rate risk of AFS portfolios would be better captured by explicitly
modelling interest rate risk in the banking book (“IRRBB”) and incorporating a rational economic
measurement of that risk as a separate Pillar 1 charge.
Currently banks in many regions would all be affected by this proposal. However, the magnitude of
the impact on any bank will depend on the extent to which it is required to keep its liquidity‐risk‐
management investments and equity investments in AFS portfolios which may in turn depend on the
relevant GAAP4 and IFRS5 rules, aspects of which are still being worked on by the FASB and IASB
respectively.
4 Generally Accepted Accounting Principles issued by the US Financial Accounting Standards Board (“FASB”)
5 International Financial Reporting Standards issued by the International Accounting Standards Board (“IASB”)
11
SignificantinconsistenciesinRWAduetodifferencesinaccountingstandardsbetweenjurisdictions:
Accounting Standards across jurisdictions differ, so that even if the valuation‐based proposal were
uniformly implemented, it could potentially produce materially different effects if there were
material differences in accounting standards, e.g. with regard to AFS portfolios. As set out in the
June 2011 ISDA paper: “It is more appropriate for bank regulation to have an appropriate boundary
than for it to have one that is completely consistent with accounting, especially given that
accounting is not itself internally consistent.”
2.4.2IssueswiththeTradingEvidenceBasedApproach
The Trading Evidence Based Approach overcomes most of the material concerns with the Valuation
Based Approach. It better aligns regulatory capital with how risks are managed by firms. Risk
management for trading books is carried out proactively, typically daily and often intra‐day. Limits
are set and monitored, Daily mark to market takes place with reporting to senior management. Risk
management occurs at the portfolio level and it follow that Trading Book classification should also
occur at this level.
However, we offer the following comments that we hope will be taken into account in developing a
Trading Evidence Based Approach:
Objectiveevidentialrequirements
We would like to work closely with the TBWG to arrive at the “objective evidential requirements” to
substantiate trading intent, and wish to highlight from the outset that an evidence based framework
should act as guidance rather than a rigid checkbox exercise, reflecting the fact that trading intent
cannot be evidenced in every case in exactly the same way.
DifficultiesinAssessingTradingLiquidity
The FRTB paper states in the paragraph requiring “objective evidence that trading instruments are
actively managed” that “banks would be required to monitor market liquidity levels (including
availability of market data)”.
As stated in Section 5 the Associations believe that assessment of the trading liquidity of instruments
should focus on the liquidity of market risk factors (i.e. on the ability to hedge) and not on the
liquidity of the instruments alone.
Standards for assessing trading liquidity face a “chicken‐and‐egg” problem. Trading liquidity initially
will be shallow for new products, broadly defined:
12
A new form of a contract on standard market factors (i.e. a new form of an option on
interest rate).
The extension of a currently traded contract to a new market factor (e.g. trading a standard
call option on a new underlying currency pair or a new equity).
The extension of a currently traded contract on a market factor to a new tenor (e.g. trading
an FX option at a longer time to expiration, or trading an interest rate swap at a longer
tenor).
If banks were to be prohibited from including any of the above transactions in their trading
portfolios, then liquidity in those markets likely would never develop. If rules of this sort had been
in place thirty years ago, important financial markets may not have developed. Many banks have
already established procedures to capture the potential risk of new market factors, for which
insufficient time series exist, by the use of prudent proxies that are subject to supervisory review.
The liquidity risk of new market factors, that are not yet highly liquid, can be captured by the use of
appropriate trading liquidity horizons, as described below. The Associations are concerned that
evidence standards should not be so rigidly defined that it would force portfolios held for trading
purposes into the Banking Book.
TransferofAssetsfromtheTradingBooktotheBankingBook
The rules governing such transfers should not be so rigid as to prevent a bank from making that
switch if it is appropriate. For example if market liquidity dries a bank may determine that the prices
of the assets in a portfolio have fallen far below their true economic value, even when measured
against stressed cumulative lifetime loss estimation. The bank may conclude that it would be
economically justified to move the assets, at their current distressed value, into the Banking Book,
with the expectation of realizing a gain in value at maturity. This could happen after, for example,
under the current proposal if the portfolio of illiquid assets has been moved already to the longest
liquidity horizon but the cost of exit still presents unacceptable losses, considering the economic
value of the portfolio.
A bank should have this option because under stressed conditions, such as occurred during 2008, the
continued loss in market value of credit sensitive portfolios will be primarily driven by increases in
the market liquidity premium rather than by objective stressed estimates of cumulative lifetime
default loss. Under such conditions it no longer makes sense to hold certain portfolios on a trading
basis.
This is consistent with the first point we made regarding the distinction between the risks in a
Trading Book as opposed to a Banking Book based on how each portfolio is managed.
To conclude, transfers from the trading book to the banking book should be authorized when
circumstances warrant and provided they are subject to strong governance and controls, and duly
documented and disclosed. Such transfers should also be subject to supervisory approval.
13
We do not believe internal audit should be first line of control to ensure positions meet the
criteria. Instead firms should rely on established control structures with the first line of control
being front office supervision, the second line of control being risk management, and the third being
internal audit.
ConcernsabouthidingtradingactivityinaliquidityAFSorHTMportfolio.
Weunderstand the regulatory concern that a bank could hide trading activities in the banking book, for example in an AFS portfolio used to manage funding liquidity risk. We also accept that where a
portfolio is under AFS, adverse market movements do feed through to CET 1. However, we do not
think the appropriate solution to that valid concern is the imposition of the Valuation Based
Approach. Other approaches should be considered to address this concern. Instead we think that
the Trading Book should properly be defined by how a portfolio is managed (Evidence Based Intent).
The Fundamental Review should include a specification of what may not be included in the Trading
Book (i.e. prohibiting certain types of transaction with certain features) and what must be included
in the Trading Book (i.e. because of how a portfolio is managed).
14
3InterestRateRiskintheBankingBook
The FRTB paper indicates that it is the intention of the BCBS to consider the timing and scope of
further work on IRRBB later this year (including possible application of a Pillar 1 capital charge for
IRRBB). However, the Associations would like to share their views on this issue. Including IRRBB
under Pillar 1 would not be easy because many of the assets and liabilities in the banking book have
indefinite maturities. We are concerned that if IRRBB were included in Pillar 1, fixed regulatory
assumptions might impose unrealistic maturity profile assumptions on indefinite maturity assets and
liabilities and on contingent assets that would not correspond to a bank’s actual experience. The
result would likely be unrealistic measures of RWA for IRRBB. We do, however, note that an IRRBB
has been imposed by the Australian authorities in Pillar I. The Associations are willing to work
closely with the BCBS on the feasibility of establishing Pillar 1 standards for measuring IRRBB
Indefinite maturity liabilities include Demand Deposit Accounts (“DDA’s”). Large banks have
developed methods for distinguishing between “core deposits”, which are stable and “non‐core”
deposits, which can run off. As a matter of illustration, one of the reasons for which DDAs' schedules
are long (and much longer than 5 years) is typically because they are coming from clients who have a
mortgage or otherwise have a long term relationship with the bank. The liabilities schedules can't be
determined without considering the global relationship with the clients holding these deposits.
These behavioural assumptions are complex and will vary widely between banks. For example a
bank suffering an idiosyncratic reduction in credit worthiness may see DDA’s fall rapidly whereas the
same bank facing the same rating downgrade may see DDA balances rise if the disruption is market
wide and it maintains or improves its rating relative to the industry (“flight to quality”).
Indefinite maturity assets include credit card receivables which roll over. Large banks have
developed methods for modelling the duration of these assets.
Other assets that create challenges in modelling are contingent credit lines, in both retail (i.e.
unused credit card lines) and wholesale (unused commitments) portfolios. Banks have historical
data on which to model these potential assets.
The Associations want to emphasize that the while we are willing to work with the BCBS on the feasibility of establishing Pillar 1 standards for IRRBB, we do not support the inclusion of IRRBB in the Trading Book framework. IRRBB should be considered independently due to the following reasons The calculation of economic capital, or RWA, for IRRBB necessarily rests on complex assumptions about the behaviour of depositors and borrowers as a function of the state of the economy, the state of the market, and other factors. The challenge of including IRRBB risk in Pillar 1 is the need to develop broad uniform principles with the flexibility to allow banks to use different assumptions, but assumptions that are capable of being validated s, based on their own experience, the characteristics of the markets they are in, the characteristics of their depositors, borrowers, and other counterparties. Given these constraints, we do not see how a standardised or uniform approach would be useful in the case of IRRBB.
Any eventual inclusion of IRRBB in Pillar 1 also should take into account the fact that this risk often
has a negative or very small positive correlation with other risk types. For example, during an e
economic downturn central banks typically lower interest rates. Banks typically have long dated
assets and shorter dated liabilities and thus are thus structured to make money when interest rates
15
fall. In addition, bank treasurers can and will use interest rate derivatives to enhance the sensitivity
of the banking book to a fall in interest rates. As a consequence, the increased net interest revenue
that results from falling rates is usually an offset to increased credit losses during an economic
downturn. Consequently a simple addition of RWA across all risk types, including IRRBB risk, would
tend to exaggerate the marginal contribution of this risk to total economic risk.
We have commented in the previous section that there is a close link between the Trading Book
boundary and the appropriate IRRBB charge.
16
4ChoiceofRiskMetricandCalibrationtoStressedConditions
4.1Introduction
In relation to Question 8 in section 4.6 of the Fundamental Review:
“What are the likely operational constraints with moving from VaR to ES, including any challenges
in delivering robust back testing, and how might these be best overcome?”
The Associations are supportive of the proposal to move from Value at Risk (“VaR”) to Expected
Shortfall (“ES”) but recognises that, for smaller or less sophisticated institutions this may present
significant infrastructure and processing burdens. ES is a coherent risk measure and is sub‐additive.
Further, because ES is the mean loss above the threshold (i.e. is an average number) while VaR is just
one point of the distribution, we would expect ES to be more stable than VaR. However, on
theoretical grounds and for practical purposes, we make a number of recommendations.
4.2AppropriatePercentile
We believe that the 95th percentile would be a more appropriate threshold from which to calculate
ES. There are three principal reasons:
a. Expected Shortfall is defined as the average of VaR‐s over various percentile levels. For
continuous distributions it equals to a conditional expected value or tail integral. There are
various ways to calculate it. One could take the average of losses exceeding the VaR level;
calculate the integral of VaR‐s of a constructed empirical distribution; fit a parametric
distribution to historical Profit and Loss (“P/L”) in order to perform the integral; or, do a
Monte Carlo simulation to generate additional loss realisations beyond the threshold.
However the simulation might itself need to make assumptions about the distribution of P/L.
The main benefit from choosing a lower percentile is that lower thresholds allow more
observations from which to form the average and therefore give a more stable and accurate
approximation. A lower percentile (e.g. the 95th percentile) would give a larger number of
excess P/L observations to average over, which improves the statistical significance of the
results and enhances their stability. By contrast, estimations beyond the 99th percentile
would be unacceptably volatile when scenarios are rolled for those using historical
simulation.
b. ES computed from a lower percentile will deliver a similar capital standard to VaR computed
at the 99th percentile for P/L tails of medium fatness. We show that Expected Shortfall with
a lower confidence level is equivalent to VaR based on the Generalised Pareto Distribution
and a range of assumptions for the tail parameter (See Appendix 1).
c. In order to maintain an ES based metric as a risk management tool it is essential that its
implied regulatory capital is not only driven by a few extreme tail events but a tail which is
representative of the firm’s business model. Choosing the threshold too high is likely to
17
result in an unstable allocation of regulatory capital to individual business lines. An
additional unintended consequence is that business lines which incur some, but not
excessive, tail risk may not attract any regulatory capital.
With respect to robustness of back testing, we believe that, while back testing the ES metric itself is
not meaningful, the established outlier back testing is important to continuously monitor model
performance. This follows because the quality of the ES metric is directly implied by the quality of
the simulated or projected P/L distribution. The present regulatory back testing, however, could be
enhanced by considering the full distribution rather than one selected percentile.
We assume that in moving to ES with tail risk properly captured, the floor to the regulatory
multiplier would be lowered to one. Appendix 1 provides some theoretical arguments for doing this.
In addition, we believe that some of the backtesting and mechanisms based on backtesting that we
are proposing to allow a smooth transition to standard rules in the case that models do not perform
as anticipated, would render a multiplier redundant.
Similarly our backtesting proposals render a floor to capital derived from internal models based on
standard rules unnecessary.
4.3DirectversusIndirectApproach
In relation to Question 5 in section 4.5 of the Fundamental Review:
“What are commenters’ views on the merits of the “direct” and “indirect” approaches to deliver
the Committee’s objectives of calibrating the framework to a period of significant financial
stress?”
The Industry welcomes the Committee’s acknowledgement of the practical difficulties surrounding
the “direct method” of identifying a suitable stress period. However we have concerns regarding the
specific example of an “indirect method” although we believe suitable alternatives exist.
The indirect method example proposed takes the form:
Where ESs is the proposed stressed measure, ESFC is Expected Shortfall based on the full set of risk
factors in the current period, ESRC is expected shortfall based on reduced set or risk factors in the
current period and MaxStressLossR is the maximum stressed loss based on the restricted set of risk
factors. One issue with this method is that ESS is not a stressed Expected Shortfall but rather a
maximum stressed loss scaled by the ratio of two expected shortfall measures. It is not even clear
that the scalar would necessarily be greater than unity but in any case it is simply not the intended
risk measure and would likely be unstable and very extreme. We believe a measure that better
captures the intent is as follows:
RC
FCRS ES
ESossMaxStressLES
18
This approach provides an Expected Shortfall measure based on current expected shortfall and
scaled by the ratio of expected shortfall based on a set of reduced risk factors scenarios observed in
a period of stress to the expected shortfall based on the same reduced set of risk factors observed in
the current period. Another possible scaling factor could be based on the ratio of standard rules in
the stressed period to standard rules in the current period. This would of course only work in the
second of the two proposals for standard rules – the “fuller risk factor approach” for revised
standard rules.
We think that both of these proposals have the following benefits:
1. Makes use of ES calibrated to current market conditions which is a better risk management
tool.
2. Avoids the calibration period being too short if we are restricted to a period of stress –
especially for longer liquidity horizons.
3. Backtesting of ES or its underlying distribution calibrated to a period of stress would not be
very meaningful in benign times.
4. ES based on a period of stress – especially with few scenarios – would be easy for firms to
arbitrage.
There are alternatives, however, to the direct and indirect methods. These include scaling up/down
scenarios to reflect stress, including extreme scenarios and the use of theoretical distributions to
estimate tail risk. There is an interesting theoretical consideration about whether P/L distributions
observed in ‘benign’ times are a result of benign volatility or simply reflect that there should be
periods of time when observations are drawn repeatedly from the body of the distribution. That is,
there is an observability issue about the distinction between time varying volatility and tail risk. Even
when observations from the body of a distribution are being observed it is still possible to calibrate
the parameters of the distribution and then extrapolate the tail to compute an expected shortfall
measure.
We also note that other capital buffers, for example Pillar 2 charges, the conservation buffer, SIFI
charge and CCAR6 already provide stressed buffers. Stressed ES in Pillar 1 is double counting, a
weakness of the current regime that regulators were seeking to remove. One could imagine an ES in
normal times for Pillar 1 and a Pillar 2 buffer based on the difference between stressed ES and non‐
stressed ES.
Calibrating ES to a stressed scenario is not really practical for a diversified portfolio where basis risk
is the main driver. A stress period calibration is suitable only for directional portfolios.
6 Comprehensive Capital Analysis and Review, Federal Reserve Board Methodology and Results for Stress Scenario Projections, March 2012
RC
RSFCS ES
ESESES
19
5LiquidityAdjustedValueatRiskorExpectedShortfall
5.1Introduction
In relation to Question 2 in section 3.3 of the Fundamental Review:
“What are commenters’ views on the likely operational constraints with the Committee’s proposed
approach to capturing market liquidity risk including the endogenous component and how might
these best be overcome?”
Section 3.3 of the Fundamental Review addresses how to factor in market liquidity. It proposes that
liquidity risk should be captured in Value at Risk (“VaR”), or Expected Shortfall (“ES”) by assigning
instruments or risk factors to one of five liquidity buckets.
The Associations support the idea of incorporating market liquidity in the Trading Book capital
framework. As we have learned from the recent financial crisis, a sudden and severe impairment of
liquidity can lead to difficulties in hedging and exiting positions, resulting in significant cumulative
mark‐to‐market losses. However, this should be done carefully because the scaling of shocks by the
square‐root of time, as proposed in the consultative paper, can have a very big effect when the
specified liquidity horizon is long. The effect on the relative contribution by risk factors of different
liquidity horizons can also be very significant. Furthermore, the incorporation of market liquidity into
VaR or expected shortfall overlaps, to a certain extent, other Fundamental Review recommendations
such as restricting the recognition of hedging and diversification benefits and stress calibration.
5.2PracticalConsiderations
The Associations see a number of practical considerations which could affect the incorporation of
market liquidity in the market risk framework.
First, one could defease risk exposures via unwinding positions, hedging underlying risk factors or a
combination of both. To encompass such possibilities, the framework should reference the liquidity
of risk factors as oppose to liquidity of instruments. The measurement of liquidity is a complex task
and it is even more complicated when we need to look at “speed” to hedge vs “speed to unwind”
and that a position can be sensitive to multiple risk factors.
Second, there is a tradeoff between the speed at which risk is being defeased and the cost to unwind
or hedge. One can choose to defease risk faster by “paying” for the ability to do so quickly. One can
unwind positions at a pace that minimizes its impact on market prices, or accumulate hedges slowly
without having to incur extra hedging cost. At the other extreme, one can also attempt to eliminate
risk exposures in a day (or less) by accepting large price concessions and/or elevated hedging cost.
20
Third, the speed at which a firm chooses to defease its risk factor exposures also depends on the
composition of the portfolio and the correlations among risk factors. Large concentrated risk factor
exposures would need more time to defease, or, a higher cost to defease if done over a short period
of time. Also, if a liquid risk factor is “hedging” other less liquid risk factors via correlation, one might
want to defease these liquid and illiquid risk factors in locked steps and with proper rebalancing to
avoid an unexpected increase in risk during the risk reduction process. Furthermore, firms might
have differential access to markets. Client relationship and other considerations will also affect a
firm’s choice of speed vs cost to defease risk.
5.3ConceptualFramework
Given the above considerations, conceptually, one can think of a general framework in which the
liquidity adjusted expected shortfall (LES) is composed of two components: A component that is
related to the forward Mark to Market (“MTM”) risk related to the liquidity horizons of its risk factor
exposures (ESh), and a component that is like an add‐on that captures hedging/unwinding cost (ESc).
LES = ESh + ESc
The ESh component can incorporate risk factor liquidity horizons that are long enough such that ESc
is minimal. Alternatively, a firm can choose to defease its risk factor exposures at a fast speed such
that ESh is small but ESc is big. In fact, a firm could even choose a uniform horizon for all its risk
factor exposures. In that case, it would have to pay for much higher hedging/unwind cost for its less
“liquid” and/or more “concentrated” risk factors. It is also important to note that while ESc could
simply be the sum of hedging/unwinding costs across risk factors, the relationship might be more
complicated as the amount one is willing to pay to defease certain risk would also depend on
whether there are other hedging or diversification benefits associated with having the particular risk
factor exposures.
Firms might find a different optimal combination of ESh and ESc. Some firms might prefer to have a
model that capture liquidity risk primarily based on ESh while others might prefer to model the
effect of liquidity risk via ESc. Some might prefer a more general setup with a combination of the
two components. Furthermore, since firms can have a different optimal combination of liquidity
horizons vs defease costs, it would be difficult to have one set of liquidity horizons or one set of
transaction costs that would work for all firms. Imposing such restrictions could lead to distortions.
However, to facilitate comparison across firms, one could compute an implied liquidity horizon from
a firm’s LES and one day ES. Specifically, if firm i has LES(i) = ESh(i) + ESc(i) and it has a one day
expected shortfall of ES1(i), then one can always compute an implied uniform liquidity horizon T(i)
such that LES(i) = ES1(i)*sqrt(T(i)). In other words, firms can have their own preferred setup and
liquidity horizons / transaction cost combinations but then can be compared via their implied
uniform liquidity horizons. Firms should stand ready to justify their calculations and their different
21
implied uniform liquidity horizons. However they should be given some flexibility in terms of how
they implement market liquidity and their estimates of liquidity horizons and hedging/transaction
cost across risk factors.
This approach is similar to option 3 discussed in Annex 4 of the Fundamental Review except that the
uniform liquidity horizon is not necessarily calculated by some prescribed weighted averaging based
on regulatory risk factor horizons. Similar to the market liquidity requirement for IRC, there should
be a uniform requirement, in principle, across jurisdictions on the capturing of market liquidity risk.
However, firms should be given flexibility on implementation.
22
6HedgingandDiversification
6.1Introduction
The Associations understand that regulators are concerned about excessive diversification in Value
at Risk (“VaR”) or Expected Shortfall (“ES”) models, but see several technical challenges with the
proposed approach. We believe that an internal model incorporating full diversification but
appropriately calibrated to stress conditions (and satisfying the enhanced validation standards
outlined in FRTB) offers a better approach for cross risk‐diversification, eliminating the issues
described above and linking capital more closely to firm’s internal risk management processes.
6.2Technicalissueswiththeproposedapproach
These include:
1. The formula specified in Equation 1 (Section 4.5.6) of the FRTB requires firms and/or
regulators to determine whether a category of risk is ‘long’ or ‘short’. It is not clear how this
could be done in practice, given a complex portfolio which may be subject to range of
complex risk factors ‐ see examples below.
2. The proposed approach requires risk factors to be classified by risk type. This may be clear
for most risk factors, but in some cases the allocation would be ambiguous – for example,
should local currency sovereign bond yields be classed as interest rate risk (as they will often
be classified for risk management purposes on non‐distressed sovereigns) or credit risk (as
would seem to be implied by recent guidance on IRC models)? Under the proposed
approach, such judgmental classification could have material impact on capital held by a
firm, for example by effectively disallowing the hedging of swaps (interest rate risk) with
local currency sovereign bonds (possibly classed as credit risk).
3. Under the proposed approach we may see examples of portfolio risk failing to be sub‐
additive, i.e. where Capital(X) + Capital(Y) < Capital (X+Y). By replacing VaR with ES as a risk
metric the FRTB would eliminate one potential cause of this undesirable effect, so it seems
unfortunate if this is re‐introduced via the diversification framework.7
4. It is not clear how cross‐risk calibrations could be calibrated – the true correlation between
potential losses in different risk types will depend heavily on the portfolio composition – see
example below. Whichever values are chosen would then imply optimal portfolio
weights/hedge ratios to minimize total capital – is it desirable for regulators to pre‐specify
these weights, regardless of portfolio composition?
7 See Marinescu & Piterbarg (2012), ‘On the non‐coherence property of the aggregate models‐based regulatory capital formula’, Barclays technical note
23
5.
6.3Pr
If, despit
in the le
‘diversif
risk valu
the dive
where α
of a firm
specify a
portfolio
would st
propose
If an app
based o
somewh
outlined
6.4De
Specific
Example
B. Presu
twice as
also take
and if so
Example
position
value ‐$
8 With α =
diversificat
A maximally
each risk cat
concentrate
than a diver
oposedA
te the additi
evel of divers
fication bene
ue. Capital co
ersification be
α would be a
m’s model of
a hard‐to‐cal
os, but instea
till require so
ed in Equatio
proach based
n risk factors
hat more sta
d above.
etailedexa
cexampleso
e 1: suppose
mably this w
s volatile as s
e into accou
o how is syste
e 2: suppose
in stock A w
98. Presuma
1 corresponding
ion.
y conservativ
tegory. Howe
ed risk catego
sified busine
Approach,
onal control
sification imp
efit’, defined
ould then be
enefit. Using
factor betw
diversificatio
librate set of
ad uses a ris
ome categor
on 1 of the Fu
d on a prescr
s rather than
ble than bet
ampleson
ofwhylong
a firm has a
would be a ‘lo
stock A, or ha
nt difference
emic risk def
a firm is trad
with value $1
ably this wou
to full modeling
ve result cou
ever this wo
ories (within
ess model.
shouldm
ls proposed i
plied by inter
as the sum o
set as the su
g the notatio
een 1 and 0
on8. This wou
f cross‐risk c
k‐sensitive p
risation of ris
undamental
ribed set of c
n trading unit
ween the lat
nHedging
g/shortcate
$100 long e
ong’ portfoli
as higher sys
e in the corre
fined?
ding basis ris
100, and a bo
uld be consid
of diversification
ld of course
uld create th
which diver
modelingn
in the Funda
rnal models,
of standalon
um of the sta
on of the Fun
set by super
uld be a mor
orrelation fa
portfolio mod
sk types, but
Review.
correlations
ts, as correla
tter, althoug
g&Diver
egorization
equity positio
o due to net
stemic risk? I
elation of the
sk between t
ought protec
dered as a ‘lo
and α = 0 corres
be achieved
he perverse
rsification ca
notbeacc
amental Revi
an alternati
ne risk by cat
andalone val
ndamental Re
rvisors accor
re natural ap
actors and lo
del, with a lim
t is seen as p
must be ado
ations betwe
gh subject to
sification
nsmaynotw
on in stock A
t $25 long po
Is this portfo
e stocks to s
two stocks, a
ction position
ong’ position
sponding to simp
by simply ta
incentive for
n be modele
ceptable
ew, regulato
ve is to com
egory, minu
lues, less som
eview,
rding to their
pproach since
ng/short cla
mit on corre
referable to
opted, it is pr
een the form
the multiple
n
work
and $75 sho
osition, but w
olio ‘long or ‘
ystemic risk
and holds a s
n in another
, even thoug
le addition of risk
aking the sum
r firms to tak
ed by firms),
ors are not co
pute the
s the fully di
me proportio
r view on the
e it does not
ssification of
lation benef
the approac
referable tha
mer would be
e technical is
ort position i
what if stock
short’? Shou
(i.e. comput
sell protectio
similar stock
gh the key ris
k by category wit
m of
ke risk in
rather
onfident
versified
on α of
e quality
t seek to
f
it. This
ch
at this be
e
ssues
in stock
B is
uld we
te beta),
on
k B with
sk to this
th no
24
portfolio is clearly the basis between the two stocks rather than directional market moves. If the
firm makes a small change to its exposure to have ‐$101 position in stock B, the portfolio would flip
from ‘long’ to ‘short’, with possibly large capital impact, even though the risk exposure (around $100
exposure to basis risk between the stocks) has hardly changed.
Example 3: suppose the firm holds a position in a derivative which results in the exposure profile
shown below – a loss if asset prices rise, a gain if asset prices fall a little, and a large loss if asset
prices fall a lot. How could firms decide if this portfolio should be considered as ‘long’ or ‘short’?
Exampleofdifficultyincalibratingcrossriskcorrelations
To illustrate how the correlation between risk categories could depend on portfolio composition in a
complex way, consider the correlation between the ‘equity’ and ‘interest rate’ categories. For spot
prices we expect the correlation to be quite low, and this is what we see in the below example –
correlation of 10‐days changes in S&P500 and 3m USD LIBOR during the 2008‐09 stress period is just
‐6%. This may lead one to set a low correlation between the Equity and Interest Rate risk classes,
reflecting the diversification one would usually expect to see. But, the correlation between changes
in implied volatility risk could be much greater – as shown below, the correlation between changes
in equity implied volatility (“vol9”) and interest rate implied vol10 over the same period is much
higher. So while ‘most’ interest rate portfolios will have low correlation to equity portfolios, even in
a stress, this would not be conservative for two portfolios driven by vega risk.
9 3 month ATM implied vol on S&P 500
10 3 month – 3 month EUR caplet implied vol
‐120
‐100
‐80
‐60
‐40
‐20
0
20
‐30% ‐20% ‐10% 0% 10% 20% 30%
P&L ($)
Return
25
Correlation of 10‐day changes,
2008‐2009 S&P 500 spot
S&P 500 3M ATM
implied vol
3M EUR LIBOR spot
3M‐3M EUR LIBOR im
plied vol
S&P 500 spot 100%
S&P 500 3M ATM implied vol ‐72% 100%
3M EUR LIBOR spot ‐6% 20% 100%
3M‐3M EUR LIBOR implied vol ‐46% 51% 46% 100%
26
7RevisedStandardisedApproach
7.1Introduction
In relation to Question 9 in section 5.3 of the Fundamental Review:
“Which of the two approaches better meets the Committee’s objectives for a revised standardised
approach?”
We believe that the ideas underlying the revised Standardised Approaches constitute a step in the
right direction as they consider key elements of modern market risk measurement, in particular
long‐/short offsetting and risk driver dependencies. This will also better align properties of the
Standardised Approaches with Internal Models based‐approaches to enable a stronger relationship
between the two.
If the Standardised Approach were to be used as a fallback approach for banks using internal
models, as proposed in the Fundamental Review, we believe that the full risk factor approach would
be appropriate for banks using internal models. The Full Risk Factor Approach is closely aligned with
banks’ internal models; hence, it would be relatively easy to implement this approach by these
banks. More importantly, the Full Risk Factor approach is more risk sensitive than the Partial Risk
Factor Approach.
It is recognized, however, that implementing the Full Risk Factor Approach may be challenging to
banks that are currently under the Standardised Approach. We recommend therefore that BCBS
seriously consider two standardised approaches – one to be used as a fallback for banks using
internal models, and another (simpler) one for banks using only the Standardised Approach. This is
consistent with the credit risk and operational risk frameworks of the Basel rules, which offer
multiple alternative approaches to banks.
7.2MandatoryCalculationofStandardisedCapitalRequirements
We agree that a mandatory calculation of standardised capital requirements for all banks would be
appropriate if the Standardised Approach were to serve as a fallback to internal models. We
propose below an approach to operationalising this that avoids the “cliff effect” expected from an
automatic shift from internal models to Standardised Approach.
However, such mandatory calculation needs to be designed in such a way not overburden internal
model banks with:
o too many parallel regulatory capital processes (e.g., in case the Basel I transitional floor
calculations are maintained) or
o high reporting frequencies (e.g., daily) for a metric which is not relevant in day‐to‐day risk
management.
27
7.3RegulatoryCapitalFloorsbasedontheStandardisedApproach
We believe the introduction of regulatory capital floors based on the standardized approach would be harmful to systemic stability, as it risks disincentivizing the use and development of internal models. If the standardized approach resulted in a capital floor that would be significantly higher than the floor as calculated under the internal models method, the standardized approach would become the binding constraint. The potential unwanted outcome would be to reduce incentives to improve banks’ internal models, which could not be good for risk management.
We believe that it is essential that firms and the regulators are incentivized to adjust and improve internal risk models over time. The core use of the standardized models should be to fuel a dialogue between banks and their regulators as to the adequacy of their current internal models and to provide regulators with improved understanding of those models and the ongoing evolution of risk in the financial system. This improved understanding can then be used to improve both internal models and standardized models. This process of ongoing improvement is best served by keeping the internal models as the clear primary tool and the standardized models as a backstop. If the standardized models implicitly become the primary risk model, then further progress in modeling risk will stall. Firms will be incentivized to optimize behavior against weakness in the standardized models rather than better modeling their actual risk.
Basing regulatory capital floors on the standardized approach would also pose challenges and extraordinary demands on the Basel Committee to make sure that the standardized approach is calibrated properly. This is because if a regulatory floor based on the standardized approach were to be implemented, the implicit assumption is that the standardized approach produces capital requirements that are more reliable and that more accurately capture the risks compared to internal models. This would require continuous validation and updating of the standardized approach. Moreover, the standardized approach is likely to be much less risk sensitive than an Internal Model‐based calculation. The resulting conflicts between risk management and capital management make it more difficult for risk managers to communicate with Front Office the need to hedge or otherwise de‐risk a portfolio, in cases where this will not lead to a reduction in allocated capital usage, which is an increasingly important constraint in assessing business performance.
28
8RelationshipbetweenStandardisedandModelsBasedApproaches
8.1Introduction
In relation to Question 3 in section 3.5 of the Fundamental Review:
“What are commenters’ views on the proposed regime to strengthen the relationship between the standardised and internal models based approaches?”
The Associations broadly agree with the aims and objectives of the BCBS to:
1. Strengthen requirements for defining the scope of portfolios that will be eligible for internal
models treatment, and
2. Strengthen the internal model standards to ensure that the output of such models reflects
the full extent of trading book risk that is relevant from a regulatory capital perspective.
8.2ApprovalCriteria
The Associations understand that regulators want to develop a methodology which allows model approval to be “turned‐off” in a more controlled way than at present. The current regime is binary with approval being full‐on or full‐off. The Industry agree that a blanket approval approach does not adequately differentiate the areas where a model works and where it doesn’t, and the consequence of turning off model approval presents a cliff effect.
To achieve this goal, the regulators have proposed two key changes to the regulatory regime. The first is to have a standardized calculation that allow for better recognition of hedging and diversification benefits – the industry welcome this change. The second is to break model approval into smaller, more discrete steps, including at the trading desk level where Profit and Loss (“P&L”) is more readily available for performance verification. This has the advantage of not only “localizing” the effect of model disapproval but also enhancing the tie between the unit at which model approval is being done and the unit at which model testing is being performed.
The Associations welcome such changes but would point out two additional aspects which are equally important towards this goal but are not discussed in the consultative paper.
8.3TheTreatmentofUnapprovedDesksorPositions
One is related to the treatment of unapproved model. In particular, whether portfolios / desks affected by the unapproved models are being excluded from overall model risk calculation in addition to standardized charges or they are still allowed to be part of the portfolio model risk calculation. The industry’s experience is that the former approach of excluding such positions from the model risk calculation can potentially create unpredictable and very punitive additional penalty when trades and corresponding hedges are not receiving model based treatment together.
This breakage of trade and hedges, creates seemingly one sided trade which can lead to very sizable model based charge not reflective of the actual risk. On this, the industry is feeling very strongly that
29
it is very important to have all components of risk included in the model based risk calculation to maintain the coherence of the model based calculation. If an addon or standardized charge is already being imposed on an unapproved model/desk, there is no reason to pull it out from the overall portfolio risk calculation. Doing so only create unbalanced treatment and distortions.
8.4APotentialApproachUsingScaling
Another important aspect to consider is related to the proportionality between penalty and model performance. Under the current regime, if a model has a high number of backtest exceptions, it might be disapproved regardless of the materiality of such exceptions. In other words, both portfolio A and B have 11 exceptions but for A, the exceptions are associated with very large losses relative to VaR while for B, the losses are just marginally beyond that predicted by VaR, then one should expect B to receive a smaller penalty than A.
A way to achieve this is to employ a addon for a model/desk that is a scaled magnitude of the standardized charge or a scaled magnitude of the difference between the standardized charge and the internal model based charge; with the scaling factor being a number between zero and one driven by model performance. For a well performing desk/model the scalar shall take a value of zero (meaning no addon). As performance worsen, the scalar shall increase towards one. This gradual penalty approach can also help avoid the cliff effect of model disapproval and make the outcome more rational. In the event that the model charge of each desk/portfolio are considered separately, this approach reduces to simply a weighted average between the internal model charge and the standardized charge.
Let IMCC(Ci) be the regulatory capital charge for a given portfolio Ci using an approved model.
Let SCC(Ci) be the regulatory capital charge for the same portfolio using a standardized approach.
Then the actual regulatory capital charge for portfolio Ci could be calculated as:
Charge(Ci) = β x SCCi + (1‐β) x IMCCi
= IMCCi + β x [ SCCi ‐ IMCCi ]
Where 0 ≤ β ≤ 1is a scaling factor set by the firm’s supervisors based on backtesting. When β = 0, full model approval results, as the model underperforms β can increase to zero at which point (assuming IMCC(Ci) < SCC(Ci)) the standardized approach is used. The transition is smooth and incentivizes firms to develop well behaved models. Note that in the equation, the capital charge for the portfolio is a weighted average of the standardized charge and the internal model based charge. I can equivalently be written as the internal model based charge plus an addon equal to a scalar times the difference between the standardized charge and the internal model based charge.
In a general setting with many sub‐portfolios, we can express the equation as:
Portfolio Charge = IMCC(whole portfolio) + �i �i [SCCi – IMCCi ]
Where SCCi is the standardized charge of sub‐portfolio i and IMCCi is the internal model charge for sub‐portfolio i.
One could also contemplate a calculation that simply utilizes the a scaled standardized charge as the addon rather than the scaled difference between the standardized charge and the internal model
30
based charge. That has the advantage of making the calculation even simpler but with some degree of double count.
8.5Eligibilityforinternalmodelsapproach
A risk factor that is eligible for modeling needs to have adequate historical data taking into account the frequency with which such data can be updated. A flexible and sensitive approach need to be taken in this determination otherwise certain risk factors would flip‐flop back and forth between modelable and non‐modelable status, due to temporary impairments in liquidity.
The Associations support an approach to move away from reliance on exception counts to a method that incorporates size of exceptions. Indeed, a high frequency of near misses should also count. At the same time, the risk metrics should be amenable to objective testing, and the Basel Committee should avoid prescribing measures that are volatile or difficult to objectively measure such as “very high percentile” or “long holding period”.
31
9CreditRiskintheTradingBook(ExcludingCVA)
9.1Introduction
The objective of this section is to capture and capitalise credit risk in the trading book which is not
already included in other risk metrics. This concept should not be confused with the Basel 2.5 notion
of “Incremental Risk”. There are a number of ways of achieving this and no single prescriptive
method is advocated. Rather the principles are set out followed by an example of one way in which
true uncapitalised credit risk (above that already captured) can be calculated. The Industry is
supportive of developing a framework which would capture spread risk within ES whilst maintaining
a separate charge for default risk.
Most of the credit risk of debt securities in the trading book can be captured by a robust simulation
in the change of credit spreads over the appropriate liquidity risk horizon. The widening of credit
spreads over the liquidity risk horizon, during stressed periods such as 2008, tends to be much larger
than would be estimated by multiplying the daily spread volatility by the square root of the number
of days during the period. This occurs because during a crisis highly leveraged firms sell assets to
meet margin calls or to preserve capital ratios. Consequently, daily changes in spreads tend to have
positive serial correlation. In contrast the square root of time calculation assumes no serial
correlation.
Let us define the incremental economic risk of a downgrade over the liquidity horizon as the
potential economic loss that is not already captured in the robust simulated widening of credit
spreads over that time period. The incremental economic risk of a downgrade will tend to be
negligible. This is because markets spreads tend to be a leading rather than a lagging indicator of
ratings changes. So long as changes in credit spreads over a liquidity horizon have been robustly
simulated, they will tend to fully include the consequence of a downgrade.
Let us define the incremental default risk over a liquidity horizon as the potential economic loss that
is not already captured by the robust simulated widening of credit spreads over that time period. A
conceptual method for doing this is described below:
The full economic risk over the liquidity horizon is the sum of two probability weighted terms: Loss
Given Default + Loss Given Non‐default
The probability weighed loss conditional on default over the liquidity horizon is equal to: :
– Loss|Default = PD*LGD
where
– PD is calculated over the liquidity horizon
– LGD has to be with respect to the current market price, not par – i.e. it is the difference
between the current market price of the security and its price given default.
32
The probability weighted loss conditional on non‐default is equal to one minus the probability of
default over the liquidity horizon, multiplied by the loss on the debt security due to the simulated
change in spread:
– Loss|Non‐Default = (1‐PD)*(loss due to spread change)
Where:
– PD is calculated over the liquidity horizon
– The loss in market value due to spread change over the liquidity horizon is with respect to the
current market value not par.
The above analysis assumes that because the debt security is in a trading portfolio it is marked‐to‐
market and has no reserve for credit loss. The current market value of the security may already be
less than par because of the widening of credit spreads that has already occurred
The incremental economic risk from default over the liquidity horizon is the difference between the
full economic risk, described above, and the loss in value that measured by the simulated widening
of credit spreads over the liquidity horizon.
The above analysis assumes that the same liquidity horizon should be used to measure the loss
conditional on the default and the loss conditional on non‐default. Under that assumption, the
incremental loss conditional on default can be taken into account by a scaling up of the change in
credit spreads over the liquidity horizon, as a function of rating, tenor and asset type.
The analysis could be extended to a different set of assumptions, i.e. if the FRTB requires banks to
calculate incremental default risk over a full year while measuring incremental spread risk only over
the appropriate liquidity horizon.
o One would have to define the total economic risk as the probability weighted sum of
the loss conditional on default anytime over the year plus the loss due to a widening
of credit spreads over the liquidity horizon, conditional on no default occurring
anytime over the year:
PD(1yr)*LGD(1yr) + (1‐PD(1yr))*Loss from credit spread(liq horizon)
o The incremental loss due to default would then be the difference between the total
economic risk, as calculated above, minus the loss over the liquidity horizon due to
credit spread widening. In principle, if this difference is relatively small compared to
the spread widening loss over the liquidity horizon, then the latter can be scaled up
(as above) to take into account the incremental risk due to default.
33
10PartialandFullerRiskFactorApproaches
10.1PartialRiskFactorApproach
The Partial Risk Factor Approach is a simple methodology in that it produces an inner product of risk weighted market values aggregated over buckets.
Strict reliance on regulatory prescribed method and parameterisation will help comparability and support level playing field
Input requirements broadly in line with COREP framework – reducing additional implementation work
However, many firms who progress beyond the simple approach will not follow this path. They will, rather, develop models more along the lines of the fuller factor approach. As such, Partial Risk Factor Approach entails a considerable degree of modeling but without the benefits of a Fuller Risk Factor Approach. Firms which already have IMM models will more naturally adopt the Fuller Risk Factor Approach.
10.2FullerRiskFactorApproach
This represents a more sophisticated approach which applies prescribed shocks to positions mapped to respective risk factors. It is similar in complexity to an internal (full revaluation) model but the somewhat simple aggregation method reduces the quality of the overall result.
Reliance on bank’s own valuation routines curbs comparability
10.3Internalmodels‐basedapproaches
The Industry supports the aim of aligning the properties of standardised approaches with internal
models‐based approaches to develop a stronger relationship between the two.
We believe that the ideas underlying the revised standardised approaches constitute a step in the
right direction as they consider key elements of modern market risk measurement, in particular
long‐/short offsetting and risk driver dependencies.
However, implementation of both approaches will require a substantial investment both at
inception (e.g., feeding all relevant information into the regulatory calculation process) and for
regular production (market data, mapping maintenance).
From a methodology viewpoint, both approaches base on simple Gaussian and Taylor assumptions,
which are not capable of reflecting complex tail risks. The bucketing is also unlikely to capture basis
risks. Hence, in order to avoid producing results well below those derived from internal models, a
conservative calibration will be essential.
It is also not clear to us how the output of the approaches shall be validated as no direct link to a loss
distribution approach and relevant time horizon is given. Outlier back‐testing (albeit not required in
the paper) is not an option here.
34
The proposed approaches are heavy on underlying assumptions (e.g., risk factor mappings) and rely
on a broad set of parameter (e.g., correlations); this will require a timely maintenance process to
adapt these models to changes in the market environment. We would like to know how policy
makers plan to achieve this.
At this stage, we are not in a position to decide on which one of the two proposed new standardised
approaches should be preferred. Such a decision should be based on a thorough quantitative impact
study following a detailed model description and an initial parameterisation proposal. Initial
comments on the approaches:
Given the significant efforts involved with implementing a new standardised approach, be it partial
or full risk factor approach, banks should also have the option to stay with the current standardised
approach framework. This is in particular relevant for banks with small trading units.
We also agree that a mandatory calculation of standardised capital requirements for all banks would
be beneficial as a uniform benchmark. However, such a requirement needs to be designed to not
overburden internal model banks with too many parallel regulatory capital processes (e.g., in case
the Basel I transitional floor calculations are maintained) or high reporting frequencies (e.g., daily)
for a metric which is not relevant in day‐to‐day risk management.
We disagree with the introduction of regulatory capital floors based on standardised approaches as
it disincentivises the use and further development of internal models and increases the difficulty of
calibrating the standardised approach. Instead, one could introduce a smooth transition between
the two (e.g., by a weighted average of both outputs depending on model performance).
35
AppenGener
We havto meetline witprovide
Assessminformaempirictrianguassumehas a ge
The tailto a con
for pararelation
with thi
If
then a q
11 Pickan
804.
131.
ndix1:ExralisedP
ve suggestedt the objectivth current lee some addit
ment of theation aboutcal evidencelate this cone that the taeneral powe
probabilitynstant of pr
ameters nship betwe
is tail. (The
denotes t
quick calcul
ds–Balkema–Balkema, A., a
Pickands, J. (1
xpectedSaretoDis
d in our respve of increa
evels. This ptional quant
relative be the shapese and leads nclusion by ail of the diser shape.11
y roportionali
and . Theeen percent
“location”
the expecte
ation show
de Haan theoand de Haan,
1975). "Statist
Shortfallstributio
ponse that cased risk senpercentile istitative supp
ehaviour of s of tails of mto the suggmaking a pstribution, c
, ty, by a gen
e value of ttile and exp
parameter
d shortfall a
ws
orem: L. (1974). "Re
tical inference
landVaRn
calibration tnsitivity by s based on pport for this
ES and VaR market distgested perceparametric acorrespondi
the probabneral power
he tail shapected short
will not
at percentil
esidual life tim
e using extrem
REquival
o a lower pusing ES, w
practice at fis choice.
at high pertributions. Pentile mentassumptioning to losse
bility of a losr law
pe parametetfall based r
need to be
le :
me at great ag
me order stati
lentsund
ercentile wowhile maintfirms, and in
rcentiles muPractice at ftioned abovn about tail ss above a s
ss greater th
er determinrisk measur
estimated
e", Annals of
stics", Annals
derthe
ould be apptaining the bn this Appen
ust be basedfirms is baseve. Here weshape, namuitable thre
han x, is giv
nes the res for distri
for our disc
Probability, 2
of Statistics, 3
propriate broad in ndix we
d on ed on
mely, we eshold,
ven, up
ibutions
cussion)
, 792–
3, 119–
36
Compar
dependi
This is u
with the
market l
Given a
neutral,
confiden
Applying
In order
governs
only if
should b
shortfall
0.9
50
0.9
55
0.9
60
0.9
65
0.9
70
ES
leve
l
ing the resul
ing only on t
seful for com
e current per
losses tend t
suitable valu
relative to t
nce correspo
g the approx
r to use this e
the value of
. Since
be chosen fro
l at
Figure 1 99% V
0
lt with the fo
he tail shape
mparing over
rcentile‐base
to follow a p
ue for the tai
he current V
onding to a ta
ximation deri
equitation w
f the momen
most financ
om the inter
.
VaR equivalent
0.1
ormula for th
e parameter
rall capital re
ed VaR meas
ower law.
il shape para
VaR measure
ail property
ived above, w
we need an es
nts of the dis
cial data show
rval , w
t ES confidence
0.2 1
96
he tail proba
:
equirements
ure, based o
ameter , w
e. For capital
of 1%, we ne
we find that
stimate of th
stribution, m
ws the 3rd‐4th
which leads to
e levels for vari
0.31/4 1/3
6.8%
96.6
bility gives u
from an exp
on the broad
we can assess
neutrality, a
eed to be
the required
he tail param
eaning the
h moment to
o a confiden
ious tail param
0.43
6%
us a useful re
pected short
ly supported
s that choice
as VaR is cali
e the solution
d confidence
meter . The
‐th mome
o still be finit
ce level for t
meters in the GP
1/2
96%
elationship,
fall capital m
d hypothesis
e of will b
brated to 99
n of
e level is g
e tail parame
nt is finite if
e, the pa
the expected
PD distribution
0.6
measure
that
e capital
9%
given by
eter also
and
rameter
d